cec - Computer Science & Engineering - University of Nevada ...

longtermagonizingInternet and Web Development

Dec 13, 2013 (3 years and 8 months ago)

83 views

Friend
Recommendations in
Social Networks

Presenter: <Name>

University of Nevada, Reno

Computer Science and Engineering Department

IEEE Congress on Evolutionary Computing, 2011

... Using Genetic Algorithms and Network Topology...

2

3

Overview


The Friend Recommendation
Problem


Goals


Network Theory


Social Genome


Link Recommendation


Experimental Results


Summary and Future Work


Questions

4

Friend Recommendations

Dynamic.

Heterogeneous.

5

Goals


Combined approach is better


Complex networks + genetic algorithms


More relevant friend recommendations


Customized user experience


Perception of friendship

6

Network Theory


Filter:


Friends of friends


Degree centrality

7

Social Genome


Facebook


Graph API


1200

users,
23K

friendships


10
-
dimensional binary string


Each bit representing social feature

8

9

Social Genome


Shared friends


Location


Age range


General interests (likes & music)


Photo Tags


Events


Groups


Movies


Education


Religion/Politics

10

Social Genome

[

1
, 0,
1
, 0, 0, 0,
1
, 0,
1
, 0
]

11

Link Recommendation

1.
Filter
: friends
-
of
-
friends, degree centrality

2a. Evolve the genome to obtain perception of
friendship

-

Pareto domination

2.
Filter
: GA with cognitive theory

2c. Compare potential against current friends

3.
Higher valuation
: if potential > current

4.
Sort potential
: best
-
> worst, take top 10

2b. Evaluate current friends

12

Results


Examined 100 Facebook users.


Randomly removed 10 friends for each individual.


Combined approach outperformed individual
counterparts.

13

Results

14

Results

15

Summary and Future Work


Friend recommendations, perception of friendship


Combined approach


Networks


Social genome


Facebook users


Future Work:


More features


Floating point representation

16

Questions?